Subject: Re: [linux-audio-dev] Interesting article on SCSI vs. IDE
From: Paul Barton-Davis (pbd_AT_Op.Net)
Date: Fri Nov 10 2000 - 01:12:38 EET
>> reset it to 8192 frames (about 170msecs), and recorded 16 tracks for
>> 40 minutes without any problems whatsoever. This is to a single Ultra
>> 2 SCSI 10K RPM disk (IBM Speedstar). Thats all I know for know.
>
>Hmm, I can get 30 MB/sec constant (not burst) data write rate to single
>10Krpm U160 SCSI disk without noticeable affect on system or other disk's
>response. (cache flush time included, with my own software)
Oh, sure. The problem only shows under "VM pressure", which arises
when one or more applications is putting the buffer cache under
considerable strain. When you're streaming a bunch of 32 bit channels
to disk, with no cache reuse whatsoever, and already have 70MB of
memory pinned down to avoid paging in the app, this pressure shows up
from time to time. In 2.4test7 and below, the VM code still can't
handle this situation properly, and ends up causing a scheduling stall
that affects the audio thread, causing an xrun.
There are many situations with intense disk i/o that would never show
this behaviour, because their interaction with the buffer cache and
VM in general is different than the recording+playback behaviour of
something like ardour. My mini disk benchmarks get the same kind of
performance that you describe (modulo a slower disk drive).
--p
This archive was generated by hypermail 2b28 : Fri Nov 10 2000 - 01:53:22 EET